Network pharmacology approaches for research of Traditional Chinese Medicines

Traditional Chinese Medicine (TCM), which plays a great role in health maintenance for the people of China, has accumulated valuable clinical experience during thousands of years of applications. It was the only medical practice in China before the early nineteenth century in the Qing dynasty. TCM is characterized by holistic and unique multi-target efficacy against chronic or complex diseases. With dedicated efforts to advance the modernization of TCM by Chinese government, Traditional Chinese Medicines (TCMs) have gained a global popularity. Pharmacodynamics material basis and effective mechanisms are two main issues to decipher the modes of action of TCMs to treat disorders and diseases ^[1-4]. For Western medicines, the conventional approach of “one disease-one target-one drug” has difficulty in providing effective therapeutic solutions for complex or multifactorial diseases, precisely because of the complex nature of diseases. Under the guidance of TCM theory, TCMs achieve holistic effects against complex diseases in a multi-component, multi-target, multi-pathway pattern, and can show satisfactory clinical results in complex diseases ^[5].

The recent decades have witnessed many new technologies that facilitate the understanding of complex interactions between TCMs and diseases ^[6]. Network pharmacology (NP) is a new discipline, which integrates systems biology, computational biology, experimental validation, and other related disciplines, serving as tools to unravel the molecular interaction network in a holistic manner. So far, NP has successfully applied to figure out the pharmacodynamics material basis and effective mechanisms of TCMs ^{[7, 8]}. Network construction and network analysis are two critical steps to unlock the underlying molecular mechanisms. There are three types of databases and tools that are frequently utilized to construct TCMs associated networks, namely compound and target related TCMSP ^[9], TCMID ^[10], ETCM ^[11], SymMap, TCM Database@Taiwan ^[12], BATMAN-TCM, PharmMapper ^[13], TargetNet ^[14], DRAR-CPI ^[15], TCM-PTD, HIT ^[16], LTM-TCM, HERB, and CPMCP; disease gene related CHD@ZJU, OMIM, DisGeNET, MalaCards, GeneCards, CTD, and TTD; and protein-protein interaction (PPI) related STRING, BioGRID, DIP ^[17], IntAct, and STITCH ^[18]. The databases constantly updated within the past three years are listed in Table 1. Nowadays, molecular biology, metabolomics, proteomics, transcriptomics, single-cell omics ^[19], artificial intelligence (AI), and other advanced theories and technologies are used in combination with NP to systematically investigate the network among herbal components, targets, pathways, and diseases. These integrations are of great importance for efficacy study, mechanism deciphering, safety evaluation, drug repurposing, and drug/drug combination design and prediction. The development and application of NP has promoted the safety, efficacy, and mechanism investigations of TCMs, which then reinforced the credibility and popularity of TCMs.

The current organ-centricity of medicine and the “one disease-one target-one drug” dogma have obstructed the understanding of complex disease and the development of effective drugs. Most of the current drugs, except for the drugs for orphan diseases, are treating symptoms rather than curing diseases. Therefore, more attentions should be paid to shift “phenotype and symptom” to “endotype and cause” in understanding and redefine current diseases. Based on the core theory of “network target”, NP proposes novel routs to investigate the modes of action of multi-components or multi-drugs synergistically acting on key network proteins to modulate the same causal disease module or signaling network ^{[5, 7]}. By this, NP focuses on curing causal mechanisms instead of treating symptoms for complex diseases ^{[20, 21]}.

However, the current application of NP in TCMs research still remain at a relatively superficial level. In the big data era, advanced theories and technologies are indispensable to improve the construction of biological/clinical-related network, and to speed up the extraction of underlying information. This review will briefly summarize the recent research progress on NP application in TCMs for efficacy research, mechanism elucidation, target prediction, safety evaluation, drug repurposing, and drug design (Fig.1).

Figure 1.  Network pharmacology based resesarch on TCMs

A biologically meaningful network reveals the systematic pharmacological modes of action of drugs, and guides the efficacy evaluation in a holistic manner. For example, a drug-disease network (DDN) is commonly consisted of drugs, drug-related targets, disease-related targets, links between the drugs and the targets, and links among the targets. The quality of databases and tools may influence the construction of networks and network analysis. Therefore, the world’s first NP evaluation method guidance‐Draft was published to standardize network pharmacological research ^[8].

Networks may be applied to model herb pairs in TCMs. By constructing the herb-ingredient-target protein networks and analyzing the multiple network-based distances (i.e., the closest, shortest, center, kernel, and separation), at both the ingredient and the target levels, the frequently used herb pairs tend to have shorter distances, while the topologically center ingredients of herbs tend to infer the modes of action of TCMs ^[39]. Metabolomics, RNA-sequencing and AI can be integrated with NP for efficacy investigation, which increases the richness of network analysis results. Metabolomics coupled with NP analysis profiles the global changes of end products in biological samples and have been used to investigate the protective effect of TCMs against ischemic stroke (IS) ^[40], depression ^[41] and myocardial ischemia (MI) ^[42]. RNA-seq technique was applied to figure out the differentially expressed genes (DEGs) to dissect the potential modes of action of TCMs ^[43]. NP-based potential target identification coupled with gut microbiota analysis was used to prevent alcoholic liver disease ^[44]. Furthermore, there are plenty of reports describing AI application in TCMs research. For example, potential biomarkers in Wolfiporia cocos were screened based on the integration of fingerprint, machine learning, and NP ^[45].

In order to solve one of the bottleneck problems in NP, i.e., how to select the key group of effective components (KGEC) from TCM formulae based on the optimal space which link pathogenic genes and drug targets, the contribution index (CI) model based on knapsack algorithm was proposed to select KGEC from disease-targets-components networks ^[46]. For drug efficacy research, molecular docking and NP were wielded to cope with the conundrums of the pharmacodynamic substances and the effect of TCMs ^[47]. In our previous study, a network-based index, Network Recovery Index for Organism Disturbed Network (NRI-ODN), was developed to measure the therapeutic efficacy of QishenYiqi formula (QSYQ) and to identify the roles of “Jun-Chen-Zuo-Shi” component herbs ^[48]. NRI was also applied to evaluate the efficacy of Shenmai injection in treating MI ^[49].

Shenmai Injection (SHENMAI), composed of red ginseng (RG) and Radix Ophiopogonis (RO), is a widely used TCM patent prescription for the treatment of MI. A MI network was constructed and microarray data from rat experiments were integrated. Then, a NP approach combined with network recovery index (NRI) was applied to quantitatively evaluate the holistic recovery rate of SHENMAI, RG and RO in the MI network. The NRI of SHENMAI, RG, and RO were 0.876, 0.494, and 0.269, respectively, which demonstrated that RG and RO exhibited synergistic effect and contributed to better protective effect of SHENMAI against MI.

Compound-target and target-pathway networks were constructed to identify the active compounds and potential targets of LJF against ALI. LPS-induced rat models and RNA-seq analysis were applied to determine DEGs and to conduct key targets verification. In this study, 28 active components, 11 key targets (CXCL2, CXCL1, CXCL6, NFKBIA, IFNG, IL6, IL17A, IL17F, IL17C, MMP9, and TNFAIP3) and IL-17 signaling pathway were achieved to show the multicomponent-multitarget effects of LJF against ALI.

As one holistic and biologically meaningful network model, NP supplies a novel research exemplification for understanding the mechanisms of action of TCMs towards complex diseases. However, the recent applications of NP in TCMs studies are still at a relatively superficial level ^[50]. With the assistance of rational combining proteomics, transcriptomics, metabolomics, AI, single-cell omics, and other advanced technologies and theories, systematical investigations of the multi-element network interactions improved the deep and holistic understanding of TCM action patterns.

The multi-compound multi-target networks are huge and complex. Therefore, it is essentail to extract critical information with high efficiency which is essential to decipher underlying mechanisms. The Random Walk Theory and Huffman-encoding algorithm were applied to detect key gene network motifs with significant (KNMS), and contribution coefficient was further calculated to uncover the mechanisms of TCM formulaes ^[51].

Proteomics was combined with NP to generate compound-inverse docking target-differential protein networks, and the therapeutic protein targets were identified for further investigation in the field of oxalate-induced renal injury ^[52].

In addition to targets at the protein level, RNA levels of target genes were also frequently investigated in NP to elucidate molecular mechanisms. Through RNA-seq or GEO datasets, DEGs were obtained from component-target-pathway networks, followed by hub targets selection, functional enrichment, and experimental validation ^[53-55].

Metabolomics was often used in combination with NP to construct drug-target-metabolite interaction networks with subsequent experimental validation to figure out mechanisms of action of TCMs ^[56-59]. Moreover, lipidomics was proposed to couple with NP, in order to provide new insights into the protective mechanisms of berberine against hyperlipidemia ^[60].

So far, single-cell omics technology, by virtue of its great capacity in resolving tissue heterogeneity, has become a revolutionary tool to deeply explore the pathogenesis of complex disease, and elucidate drug mechanisms of action. Single-cell omics technology presents great potential in the discovery of pharmacodynamic substances, construction of action networks, and elucidation of integrated regulatory mechanisms, which bring new opportunities for modern research in TCMs ^[19]. Compared with traditional sequencing methods, single cell sequencing can result in more precise cell subsets with a higher resolution to reveal the dynamic regulation of TCMs on different cell populations. In our previous work, single-cell transcriptome sequencing was utilized to sequence CD45⁺ cells in the heart of MI mice after intervention with tanshinone ⅡA for 3, 7, and 14 days. The dynamic regulation of different immune cells on post-MI damage repair were clarified, and the specific mechanisms of tanshinone ⅡA in the treatment of MI were further interpreted, providing a scientific basis for its clinical application ^[61].

DPP4 was specifically upregulated in diabetic kidney disease (DKD) podocytes through single-cell sequencing analysis, and further PPI network and enrichment analysis of differential genes associated with DPP4 in DKD podocytes was conducted ^[62]. The single cell RNA sequencing data of human lungs were integrated with NP to elucidate the certain cell and target genes related mechanisms of anti-COVID-19 herbal medicine ^[63], while the combination of single-cell sequencing and NP was employed to evaluate the effect of natural products on promoting hair regeneration by activating the FGF pathway in dermal papilla cells ^[64].

In addition ^{[19, 65]}, spatial transcriptome sequencing can obtain the spatial information of drug action on target tissues, which is helpful to deeply analyze the spatial heterogeneity of the action of TCM components on target cells, in order to reveal its global and synergistic mechanism. Single-cell multimodal omics can describe the life activities of cells at different levels, and provide a more comprehensive, systematic, and holistic understanding of the complex process of how Chinese medicines play a curative effect.

Nowadays, AI-based deep learning and cloud computing application in the field of traditional medicines provide new ideas for the studies of new targets and modes of action, promoting the modernization of TCMs ^[66-68]. An intelligent formula recommendation system based on deep learning (FordNet), fusing the information of phenotype and molecule, was proposed based on a deep neural network based quantitative optimization model ^[69]. AI was applied for material basis investigation, as a highly efficient system for screening the hypoglycemic constituents of Xiaokewan was developed with integration of intelligent recognition technologies in mass spectrometry dataset and computerized NP ^[70]. Recently, AI AlphFold2 has been successfully applied to predict the protein structures of therapeutic targets obtained from PPI network analysis, which predicts the mechanism of celastrol for the treatment of type 2 diabetes ^[71].

Furthermore, the underlying mechanisms of TCM formulae in the treatment COVID-19 remain unclear. NP coupled with other advanced technologies is a promising approach to elucidate the modes of action of TCMs against COVID-19 ^{[72, 73]}.

Single cell RNA-sequencing (scRNA-seq) was applied to define the cell types in the skin responding to treatment with hair-regeneration supplement SBM, including hair follicle stem cells (HFCs) and dermal papilla cells (DPCs). Intercellular communication networks of the FGF ligands/receptors were constructed and the symbol gene Fgf7 in DPCs was upregulated with SBM treatment. UPLC/MS and UPLC/LTQ-Orbitrap-MS were used to detect the main components enriched in the skin after SBM smearing, which indicated that epigallocatechin gallate and stilbene glycoside were useful to promote new hair growth. NP studies were further conducted to investigate the main components of SBM for modulating seborrheic alopecia, and PPI analysis revealed that FGF7 acted an important role in promoting new hair growth with SBM.

FordNet is a TCM formula recommendation system developed with integration of AI and TCM NP theory. A deep neural network based quantitative optimization model was developed for TCM formula recommendation. Convolution neural network and network embedding were applied to extract the feature of diagnosis description and the feature of TCM formulae, respectively. As a result, the intelligent recommendation system for TCM formulae performed obviously better than baseline methods, where the molecular information facilitated FordNet to improve the hit ratio by 17.3% compared with the only-macro-information used model. The proposed system showed potential to improve clinical diagnosis and treatment, and help to improve TCM NP development.

Metabolomics is commonly employed to evaluate the effectiveness and toxicity of drug treatment by non-discriminatory monitoring of metabolites during administration. NP combined with metabolomics exhibited a powerful solution to investigate the herbal toxicity and related mechanisms, and this approach was applied to uncover the protective role and detoxification mechanism of Yunnan Baiyao (YNBY) formula ^[74]. As toxic herbs are commonly prescribed in formulae together with other herbs to contribute to the entire therapeutic effects, the mechanisms and toxic substances causing liver injury are usually unclear, NP and in vitro experiments were integrated to determine the potential protein targets and molecular mechanisms of mitochondrial toxicity of Xian-Ling-Gu-Bao (XLGB) oral preparation ^[75]. Our research group proposed the concept and framework of network toxicology and forecasted its application in TCMs research ^[76], which can comprehensively discover the potentially toxic substances and explain toxicity mechanisms involved. Furthermore, in-depth studies of toxicity Q-markers were conducted to provide the material basis and technical support for the safety evaluation of TCMs ^[77].

The liver injury induced by XLGB, the drug for the treatment of osteoporosis in China, remains high, but the mechanisms and toxic substances are still unclear. Targets associated with mitochondrial toxicity, compounds and its potential targets were integrated to perform NP analysis, in order to uncover the potential mechanisms of XLGB induced hepatotoxicity. Seahorse assay was conducted to evaluate mitochondrial function. Further experimental studies showed the expression of mTOR, p-mTOR, Raptor, PI3K, Beclin 1, ATG5 and Caspase-9 were up-regulated and the expression of Bcl-2 was down-regulated, suggesting that the PI3K/mTOR signaling pathway and mitochondrial apoptosis may attribute to the hepatotoxicity induced by XLGB.

Caowu (CW, Aconiti Kusnezoffii Radix) is one herbal in YNBY, whose detoxification mechanism is unclear. NP method was used to predict the targets and pathways of CW. SD rats in different drug groups were used for tissue histopathology examination and urine sample metabolomics analysis to evaluate CW toxicity. Finally, 44 potential toxicity biomarkers and 14 pathways were identified to be involved in the toxicity of CW. Lysine degradation was the core metabolic pathway of detoxification of YNBY. Through integration of NP results, targets including ACHE, SLC6A3 and SLC6A4 may contribute to the protective role of other herbs in YNBY.

Drug repurposing (or drug repositioning) is an emerging concept to employ registered drugs for new indications. It is undeniable that during the COVID-19 pandemic, TCMs played a critical role in saving people’s lives, in which drug repurposing and NP manifested good values to identify herbal ingredients for obtaining anti-COVID-19 potential ^[78]. Ingredients from TCMs are often characterized by unique polypharmacologic properties ^[79]. For repurposing the anti-IS ingredient butylphthalide against central nervous system (CNS) diseases, a strategy was addressed based on NP and molecular docking integration ^[80]. Recently, a NP investigation platform SmartGraph has been released which can elucidate mechanism-of-action, conduct drug-repurposing, and predict off-target effect ^[81]. NeDRex, a generically applicable integrated platform for network-based drug repurposing and disease module discovery, can be employed to construct heterogeneous biological networks, in order to obtain disease modules containing potential drug targets ^[82]. An integrated drug similarity network (iDSN) in a proposed systematic framework PIMD was applied to predict drug repurposing, with better performance than other DSNs ^[83].

Troxerutin is used as anticoagulant and thrombolytic agent in clinical practice. Bioinformatics and NP were used in combination to find that troxerutin had potential to improve UC. In-depth in vitro and in vivo experiments presented the anti-oxidative stress and anti-inflammation effects of troxerutin, which may be associated with its network regulation of signaling pathways in tight junction barrier function, mitochondrial morphology, oxidative stress, cell apoptosis, tissue fibrosis and inflammatory cell infiltration.

Different network distance metrics (Z_dc, $ {I}_{DSD}^{min} $) were proposed to quantify the association between drug targets and COVID-19 disease targets in the PPI network. A drug with Z_dc< 0 or a smaller $ {I}_{DSD}^{min} $ indicates its potential as an anti-COVID-19 candidate. Then, bioactive herbal ingredients and chemicals with anti-COVID-19 effects were predicted. Molecular docking and gene enrichment analysis were applied to illustrate the potential mechanisms of the candidates. Rosmarinic acid, kaempferol and fingolimod might be multi-target inhibitors to inhibit 3CLpro and ACE2 to hinder COVID-19 progression.

NP has been recognized as the next paradigm for drug-discovery ^[85]. TCMs are considered as the promising reservoir for drug discovery. In accordance with TCMs in a holistic therapeutic mode, multi-compound drug candidates can be screened and further modified to exert beneficial synergistic effects for the treatment of multifactorial diseases.

Nowadays, the design and discovery of multi-targeted or multi-component drugs are still challenging. Notably, in silico methods, i.e., inverse docking, pharmacophore modeling, machine learning methods and approaches derived from NP show good value for new drug design ^[86]. For IS, there are no effective drugs, and the multifactorial properties, multi-targeted or multi-compound combinatorial drugs are needed. Nox4 is a primary causal therapeutic target for IS, and protein-metabolite network analysis, i.g., guilt by association, was useful to predict and pair synergistic mechanistic disease targets for further drug discovery ^[21]. NP was employed to analyze the active phytochemicals of Cnidium monnieri in the treatment of hepatocellular carcinoma (HCC), which provide evidence for developing anti-HCC drugs based on active phytochemicals ^[87]. Natural Product-based Drug Combination and Its Disease-specific Molecular Regulation (NPCDR) was developed to facilitate natural product-based drug combination design ^[88]. INPUT is an intelligent NP platform developed to facilitate the researches on drug discovery ^[89]. Computational methods that fused feature projection fuzzy classification and super cluster classification were employed to identify drug-target interactions for drug discovery and development ^[90]. From drug-protein heterogeneous network, effective drug combinations were predicted by a gradient tree boosting classifier, which is a systematic and reliable approach for drug discovery ^[91].

UC is an intractably disease, whose etiology is not completely understood. Artemisinin is an active component extracted from Artemisia annua with pleiotropic effects, and exhibited potential to treat UC, but the underlying mechanisms reamin unclear. Based on NP related databases, the interaction network of artemisinin and UC was constructed, and then network analysis was performed, in order to reveal the mechanisms and critical targets of regulation. Molecular docking results showed the highest affinity between GALNT2 and artemisinin, and GALNT2 was pointed as the receptor to carry our de novo hits design. Four candidate compounds showed tighter interaction with GALNT2. Molecular dynamics simulation demonstrated that three of the four designed compounds stably bound to the receptor pocket of GALNT2. Thus, NP and AI technologies provide promising routes for natural-active-components-based drug design for the treatment of diseases.

Compared with single targeted therapy, multi-targeted or pathway-targeted treatment for mCRC showed improved efficacy. CADD methods were proposed to explorer the binding modes of apigenin and kaempferol with EGFR/HER2 and MEK targets. Furthermore, NP methods were applied to construct and analyze a compound-disease-target network. Gene ontology enrichment and KEGG pathways were analyzed to explain the molecular mechanisms of actions. The results showed that apigenin and kaempferol were the potential inhibitors of EGFR, HER2, and MEK1. Finally, an in vivo CAM assay confirmed the anti-angiogenic potential of apigenin and kaempferol.

With the advent of advanced and intelligent technologies in the past two decades, NP has been improved and deely implemented, and presented a great potential as the next drug-discovery paradigm. NP is developed to elucidate causal mechanisms instead of treatment of symptoms. The “component-target-pathway” network of traditional NP study still relies on traditional omics experimental data, which however are unable to echo the “treatment based on syndrome differentiation” theory of TCM. The constructed network model cannot exactly represent the complex interactions between the pharmacodynamic substances and the biological regulatory networks, which limits the improvement of efficiency and accuracy of network analysis. Single-cell omics provides more accurate and high-resolution experimental data for invetigating the targets of TCMs. It can further boost the construction of the “component-target-pathway” spatial regulatory network, and facilitate the development of precision medicine with TCM characteristics. On the other side, AI algorithms utilize different types of data based on various strategies to complete multiple tasks, e.g., search and discrimination, which is really suitable for solving the issues concerning big data analysis through NP studies. In a word, “NP is not only pure NP; NP needs rational coupling with the advanced and intelligent technologies to be the engine of TCM modernization”.

In the future, several topics are still challenging and more efforts should be devoted to improve TCM NP. High quality and standardized databases consisting of TCM clinical practice experience, TCM formulae, compounds, targets, diseases/disease genes, pathways, RNAs, exosomes, and organ-organ/cell-cell/organelle-organelle communications are critical, which can lay a solid foundation for future NP studies. The network is multi-level and dynamic, representing biological meanings with real-time displaying of organism status at certain timepoint or under certain interference. With the assistance of AI, the network can “speak” with data and graph output. For target validation, it should not only limit to investigate the direct interaction between targets and active ligand compounds, but also look into the indirect targets and even off-targets the effects of potential compounds, which may contribute to a holistic understanding of TCM modes of action. The compatibility rule of TCMs should be investigated at the single cell and single molecular level, which can disclose the nature of TCM therapy and disease curation. Through mechanism deciphering at a high-resolution level, i.e., from organ regulation to single cell and molecule in a dynamic pattern, the new paradigm of drug efficacy and safety evaluation will facilitate the precision practice of TCM and boost TCMs/natural medicine-oriented drug repositioning and drug discovery in the new era.